NVIDIA model product
Nemotron 3 Ultra is NVIDIA's frontier-scale open model with 550B total / 55B active parameters, built for agentic reasoning, long-context analysis, tool use, and high-stakes RAG. It uses a hybrid Latent Mixture-of-Experts (LatentMoE) architecture interleaving Mamba-2, MoE, and select Attention layers, with Multi-Token Prediction (MTP) for native speculative decoding, and is pre-trained on ~20T tokens with an NVFP4 recipe. Reasoning is configurable on/off (plus a medium-effort mode) via the chat template. It supports up to a 1M-token context and 10 languages (English, French, Spanish, Italian, German, Japanese, Hindi, Korean, Brazilian Portuguese, Chinese). Released with open weights, training data, and recipes under the OpenMDW-1.1 license.
Updated Aug 10, 2026. Default version: Nemotron 3 Ultra (550B A55B)
Structured fields from the published default version.
This profile uses the latest version under this unique model that has a calculated LLMBoard score. Arena and price are excluded.
Published benchmark records for the scored version Nemotron 3 Ultra (550B A55B).
| Apex | 0.8 | 1 | 2 | 100.0% | C | |
| IMO-AnswerBench | 0.9 | 1 | 19 | 100.0% | C | |
| OmniScience | 0.8 | 1 | 2 | 100.0% | C | |
| PinchBench | 0.9 | 1 | 4 | 100.0% | C | |
| ProfBench | 0.6 | 1 | 1 | 100.0% | C | |
| RULER | 0.9 | 1 | 4 | 100.0% | C | |
| IFBench | 0.8 | 2 | 28 | 96.3% | C | |
| GDPval | 0.5 | 3 | 3 | 0.0% | C | |
| CritPT | 0.0 | 4 | 4 | 0.0% | C | |
| LiveCodeBench v6 | 0.9 | 4 | 53 | 94.2% | C | |
| LongBench v2 | 0.6 | 4 | 17 | 81.3% | C | |
| MMLU-ProX | 0.8 | 5 | 32 | 87.1% | C | |
| TAU3-Bench | 0.2 | 5 | 5 | 0.0% | C | |
| Multi-Challenge | 0.6 | 6 | 29 | 82.1% | C | |
| WMT24++ | 0.8 | 6 | 23 | 77.3% | C | |
| AA-LCR | 0.7 | 8 | 15 | 50.0% | C | |
| Finance Agent | 0.5 | 8 | 8 | 0.0% | C | |
| MMLU-Pro | 0.9 | 9 | 129 | 93.8% | C | |
| SciCode | 0.4 | 9 | 18 | 52.9% | C | |
| Terminal-Bench 2.1 | 0.6 | 16 | 17 | 6.3% | C | |
| Finance Agent v2 | 0.4 | 20 | 26 | 24.0% | B | |
| SWE-bench Multilingual | 0.7 | 21 | 34 | 39.4% | C | |
| Humanity's Last Exam | 0.4 | 37 | 92 | 60.4% | C | |
| GPQA | 0.9 | 42 | 233 | 82.3% | C | |
| BrowseComp | 0.4 | 49 | 58 | 15.8% | C | |
| SWE-Bench Verified | 0.7 | 56 | 104 | 46.6% | C |
Preference and agent-evaluation signals from published Arena datasets.
The default version has no published Arena rows, or its source alias has not been resolved.
Official vendor API PAYG pricing is summarized first. The table then lists individual provider offerings without treating their minimum as the official price.
| Kenari | nemotron-3-ultra-550b-a55b | global | N/A | N/A | 1M |
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
All published versions linked to this unique model. The score columns identify the version used by the current overall ranking.
| Nemotron 3 Ultra (550B A55B) | 72.1 | 550B | 1M | 128K | Yes | OpenMDW License v1.1 |
A concise description based on the published model registry.
Nemotron 3 Ultra is NVIDIA's frontier-scale open model with 550B total / 55B active parameters, built for agentic reasoning, long-context analysis, tool use, and high-stakes RAG. It uses a hybrid Latent Mixture-of-Experts (LatentMoE) architecture interleaving Mamba-2, MoE, and select Attention layers, with Multi-Token Prediction (MTP) for native speculative decoding, and is pre-trained on ~20T tokens with an NVFP4 recipe. Reasoning is configurable on/off (plus a medium-effort mode) via the chat template. It supports up to a 1M-token context and 10 languages (English, French, Spanish, Italian, German, Japanese, Hindi, Korean, Brazilian Portuguese, Chinese). Released with open weights, training data, and recipes under the OpenMDW-1.1 license.
Use the benchmark, Arena and pricing sections above as separate evidence. A missing field means the current data snapshot does not support that claim.
Data snapshot: 2026-08-07. Editorial model content is not available in the backend.
Open a comparison with the three ranked models immediately above and below this model.
Recommendations prioritize the same model type and family, then the closest published LLMBoard score.
Common questions about Nemotron 3 Ultra.
Nemotron 3 Ultra's default version was released on Jun 4, 2026.
No official standard PAYG price is currently available for Nemotron 3 Ultra.
Nemotron 3 Ultra is published under NVIDIA in the model registry.
The default version has a 1M token context window.
Yes. The default version is marked as open weight under OpenMDW License v1.1.
1 published provider offerings are linked to the default version.
Nearby ranked alternatives include Gemini 3 Flash, Qwen3.6 Plus, Qwen3.6 27B.